Debate on whether current LLMs can formulate math conjectures better than humans
tobyordoxford · x · 2026-08-24
Toby Ord and David Manheim discuss the capability of current LLMs to generate mathematical questions or conjectures. Ord argues that human mathematicians add substantial information through their selection, unlike random propositions. He notes that while quality is uneven, there is currently little evidence in either direction regarding LLM performance relative to humans.
More from AGI Musings
- AI abundance and the demise of the 2% inflation target — seldondev · 2026-08-24
- NotebookLM questions can be pasted, making AI homework hard to circumvent — _akpiper · 2026-08-24
- NotebookLM gives questions to ask, removing all thought from learning — _akpiper · 2026-08-24
- Opinion: AI agents are useless gimmicks for non-coders — SEND_ME_YOUR_ASSPICS · 2026-08-24
- Autonomic vs. Automated Security: Intent and Incentives in AI Defense — philvenables · 2026-08-24
- Opinion: Ambient compute will supersede phones like phones did desktops — curious_vii · 2026-08-24